Segment the person
A consistent person mask isolates the subject throughout the shot.
EFFECT / CUTOUT COLLAGE
Tracking Cutout Collage Video Effect
Remove the live person from the background and rebuild their walk as overlapping, scene-tracked paper scraps. New frozen poses appear step by step while the original background keeps moving at normal speed.
RUN THE EFFECT
Build a cutout trail from one hosted video.
Paste a public MP4 URL. The beta works best with one visible person in a single continuous vertical shot.
HOW IT WORKS
A motion trail built from the real footage.
The model isolates the person, inpaints a clean moving background, samples torn-paper cutouts, then tracks every scrap against its own nearby background features so depth, parallax, and scale respond independently to the camera.
Sample movement
A new frozen cutout is captured only after the subject moves far enough.
Anchor the trail
Each cutout keeps a foot anchor and its own local background feature track, creating depth-aware parallax during camera movement.
Layer the collage
Overlapping paper cutouts replace the live person while the clean background and source audio continue at normal speed.
Real frames. Graphic echoes.
No replacement images are generated. Every silhouette comes directly from the uploaded footage, with adjustable spacing, lifetime, density, and edge treatment.
USE CASES
Made for movement-led edits.
Music videos
Turn a walk, dance, or performance into a rhythmic trail of held poses.
Fashion films
Build editorial silhouettes around runway movement and outfit reveals.
Short-form edits
Create a readable visual hook that lands quickly in vertical video.
Campaign systems
Reuse one tracked collage treatment across talent, products, and formats.
EFFECT STUDIES
Ten rhythms for one movement.
6 / 1.8fps / 7px3 / 2fps / 12px6 / 2.8fps / 8px10 / 4fps / 4px4 / 2.2fps / 18px8 / 5fps / 10px5 / 3fps / 0px6 / 3.6fps / 10px5 / 2.6fps / 7px12 / 5.5fps / 6pxRun the collage as one API job.
The BLXSSXM endpoint keeps provider credentials server-side, starts the renderer, and returns a prediction id for automatic or manual polling.
POST /api/effects/tracking-cutout-collage
{
"personVideoUrl": "https://.../walk.mp4",
"maxEchoes": 6,
"spawnDistance": 0.075,
"outlineWidth": 7
}FAQ
Before you render.
Does the effect generate new images of the person?+
No. Every cutout is sampled from the original video and combined with its segmentation mask and a generated outline.
Are the cutouts tracked in 3D?+
The uploaded-video solver tracks each cutout against a separate local scene plane. This produces 2.5D parallax and perspective-aware scale without LiDAR. Hard cuts, severe blur, or featureless backgrounds can break the solve.
What kind of input works best?+
Use one continuous 3–10 second shot with one prominent person, clear body contrast, and no cuts. This public runner currently accepts a hosted MP4 URL.
Can I call it from another app?+
Yes. POST a hosted video URL and settings to /api/effects/tracking-cutout-collage, then poll the returned prediction id.
What can reduce quality?+
Heavy occlusion, motion blur, low contrast, multiple overlapping people, and large sideways camera moves can make segmentation or anchoring less stable.